Skilled migration: the perspective ofdeveloping countries¤

Frédéric Docquiera;b and Hillel Rapoporta;c;da CADRE, University of Lille 2, France

b IZA, Institute for the Study of Labor, Germanyc Department of Economics, Bar Ilan University, Israel

d Stanford Center for International Development, Stanford University, USA

Abstract

In this paper we focus on the consequences of skilled migration for source(developing) countries. We …rst present new evidence on the magnitude of the"brain drain” at the international level and then discuss its direct and indirecte¤ects on human capital formation in developing countries in a uni…ed stylizedmodel. Finally we turn to policy implications, with emphasis on migration andeducation policy in a context of international migration.

¤The …rst author thanks the World Bank for …nancial support (Contract PO. 7620076 - UPI269656). We thank Arye Hillman and Maurice Schi¤ for helpful comments. The usual disclaimerapplies. Corresponding author: Frederic Docquier, CADRE, University of Lille 2, 1 Place Deliot,59024 Lille, France. Email: f.docquier@skynet.be.

1

NON-TECHNICAL SUMMARY In this paper we focus on the consequences of skilled migration for

developing countries. We first summarize the recent findings of Docquier and Marfouk (2004) on the international mobility of the highly-skilled; their estimates, based on immigration data collected from nearly all OECD countries in 1990 and 2000, show that the “brain drain” has gained in magnitude over the period covered, but that substantial differences remain across regions and income groups.

The central section presents the theoretical arguments of the ''new'' and ''old'' brain drain literatures in a fully harmonized framework. We first review the early brain drain literature, which tends to emphasize the losses to those left behind. We then ask whether the traditional detrimental effects stressed in the early literature may be offset by potentially beneficial effects emphasized in more recent contributions (remittances, return migration, creation of trade and business networks, and possible incentive effects of migration prospects on human capital formation at home), and provide empirical evidence on these different channels where available. According to most existing studies, it is unlikely that remittances, return migration or other ways through which highly-skilled emigrants continue to impact on their home country's economy are significant enough to compensate sending countries for the losses induced by the brain drain. By contrast, cross-country comparisons provide supportive evidence that migration prospects foster domestic enrollment in education. Additional investigations are clearly needed to assess the net effect of emigration on human capital formation at home.

The last section explores some of the policy implications of the analysis, with emphasis on migration policy and education policy. We show that from the perspective of developing countries, the "optimal" emigration rate for the highly-skilled is zero where liquidity constraints are strongly binding and then conforms to an inverse-U shape pattern with respect to the country's level of development. As to education policy, we discuss the role of taxes and education subsidies when human capital is mobile internationally and show how optimal policy responses to such mobility are affected by the parameters of the model.

Contents1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 How big is the brain drain? . . . . . . . . . . . . . . . . . . . . . . . . . . 53 Theory and evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133.1 The model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133.2 The traditional view . . . . . . . . . . . . . . . . . . . . . . . . . . . 153.3 Temporary migration . . . . . . . . . . . . . . . . . . . . . . . . . . . 183.4 Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213.5 Remittances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243.6 Network e¤ects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

4 Policy implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284.1 The optimal rate of migration . . . . . . . . . . . . . . . . . . . . . . 294.2 Education policy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2

1 IntroductionThe current wave of economic globalization has opened a window of opportunity forhuman capital to agglomerate where it is already abundant and yet best rewarded,i.e., in the most economically advanced countries. This natural tendency has beenstrengthened by the gradual introduction of selective immigration policies in manyOECD countries since the 1980s. What started as an e¤ort to increase the "qual-ity” of immigration in countries such as Australia or Canada has developed into aninternational competition for attracting the highly educated and skilled. Togetherwith traditional self-selection e¤ects on the supply-side, this explains the overall ten-dency for migration rates to be much higher for the highly-skilled. While the worldexport/GDP ratio has increased by 51 percentage points between 1990 and 2000(WTO, 2004), the total number of foreign-born individuals residing in OECD coun-tries has increased in the same proportion (51%) over that period, a …gure that jumpsto 70% for highly-skilled migrants against only about 28% for low-skilled migrants(Docquier and Marfouk, 2004).What are the consequences of this human capital ‡ight for sending (developing)

countries? In a world of perfect competition with complete markets, the free mobilityof labor would seem to be Pareto-improving: migrants receive higher incomes, nativesin the receiving countries can share the immigration surplus, and remaining residentsin the sending countries can bene…t from the rise in the land/labor and capital/laborratios. However, in the case of highly-skilled migrants, such labor movements alsogenerate a number of ”externalities” that have to be factored in. First, skilled mi-grants are net contributors to the government budget and their departure, therefore,increases the …scal burden on those left behind (…scal externality). Second, skilledlabor and unskilled labor complement one another in the production process; in acontext of scarcity of skilled labor and abundant unskilled labor, as is the case indeveloping countries, skilled labor migration may have a substantial negative im-pact on low-skilled workers’ productivity and wages (intragenerational spillover) andincrease domestic inequality. Third, human capital depletion through emigrationwould seem to impact negatively on a country’s growth prospects, inasmuch as hu-man capital formation is now viewed as a central engine of growth (intergenerationalspillover). Fourth, as demonstrated in various new economic geography frameworks(e.g., Fujita et al., 1999), skilled labor is instrumental in attracting FDI and fosteringR&D expenditures (technological externality); hence, the mobility of human capitalis contributing to the concentration of economic activities in speci…c locations, at theexpense of origin regions.On the other hand, high-skill migration may also induce positive feedback e¤ects

as skilled emigrants continue to a¤ect the economy of their origin country. Suchpossible feedbacks include migrants’ remittances, return migration after additionalskills have been acquired abroad, and the creation of networks that facilitate trade,capital ‡ows and knowledge di¤usion.

3

Given the many channels involved, an evaluation of the exact impact of the mi-gration of skilled labor (the ”brain drain”) for source countries is a very complex task.As we shall advocate in this paper, most of this impact may ultimately be capturedthrough the e¤ect of emigration on the composition of the labor force, that is, on thestock of human capital per worker remaining in the home country. Until recently,empirical attempts in this direction have been hampered by the lack of harmonizedinternational data on migration by origin country and education level. In the absenceof such empirical material, the debate has remained almost exclusively theoretical.The early ”brain drain” literature of the 1970s emphasized its negative consequencesfor those left behind. The main conclusions were that skilled emigration contributesto increased inequality at the international level, with the rich countries getting richerat the expense of the poorer countries. By contrast, more recent contributions askwhether the traditional negative e¤ects of the brain drain stressed in the early lit-erature may be o¤set by possible bene…cial e¤ects arising from remittances, returnmigration, creation of trade and business networks, and possible incentive e¤ects ofmigration prospects on human capital formation at home.1 In particular, a new braindrain theoretical literature has emerged around the idea that migration prospects maywell foster human capital formation in developing countries even after actual emigra-tion is netted out; this literature studies the theoretical conditions under which theoverall e¤ect of the brain drain may be positive (i.e. brain drain reduces internationalinequality). During the last two decades, there has been a signi…cant increase in themagnitude of the brain drain. However, as recent theories show, it could be thatsome developing countries have experienced a social gain from this brain drain.We …rst summarize in Section 2 the data on the magnitude of the brain drain, and

then provide new estimates on the international mobility of the highly skilled; ourmeasures are based on immigration data collected from nearly all OECD countriesfor 1990 and 2000 by Docquier and Marfouk (2004). These data show that the braindrain has gained in magnitude over the period covered although substantial di¤erencesremain across countries and regions. Section 3 presents the theoretical arguments ofthe ”new” and ”old” brain drain literatures in a fully harmonized framework: we …rstreview the early brain drain literature, and contrast it to more recent models. Wealso review the various channels whereby skilled migrants may impact on their homecountry after they have left (remittances, return migration, networks), and provideevidence on these di¤erent channels when available. Section 4 is dedicated to policyimplications, with emphasis on migration policy and education policy in a context ofinternational migration. Section 5 concludes.

1Grubel and Scott (1969) already mentioned the various possible feedback e¤ects of the braindrain for source countries (remittances, networks, innovation in the host country that may spilloverto the origin country, etc), and argued that the short-term loss (due to intragenerational and …scalexternalities) could well be o¤set in the the long run. However, such possible feedbacks were con-sidered too small to make a di¤erence, and the brain drain literature of the 1970s focused on itsdetrimental (short-term) impact.

4

2 How big is the brain drain?There is clear evidence that the brain drain has increased dramatically since the1970s, both in absolute and relative terms. Nearly 30 years ago, the United Nationsestimated the total number of highly-skilled South-North migrants for 1961-72 atonly 300,000 (UNCTAD, 1975); less than a generation later, in 1990, the U.S. Cen-sus revealed that there were more than 2.5 million highly educated immigrants fromdeveloping countries residing in the U.S. alone, excluding people under the age of25 (that is, without counting most foreign students). Country studies commissionedby the International Labor Organization also show that nearly 40% of the Philip-pines’ emigrants are college educated, and, more surprisingly, that Mexico in 1990was the world’s third largest exporter of college-educated migrants (Lowell and Find-lay, 2001). Since 1990, the chief causes of the brain drain have gained in strength andthese increasing trends have been con…rmed. Indeed, selective immigration policies…rst introduced in Australia and Canada in the 1980s have spread to other OECDcountries: the Immigration Act of 1990 as well as the substantial relaxation of thequota for highly skilled professionals (H1-B visas) in the US constitute a very signi…-cant change in the immigration policy of the country that remains the destination ofabout 40% of the immigrants to the OECD area. Finally, a growing number of EUcountries (including France, Germany, Ireland and the UK) have recently introducedsimilar programs aiming at attracting a quali…ed workforce (e.g., through the creationof labor-shortage occupation lists) (OECD, 2002).Until very recently, there were no comparative data on the magnitude of the

brain drain. The …rst serious e¤ort to put together harmonized international data onmigration rates by education level is due to William Carrington and Enrica Detra-giache from the International Monetary Fund, who used US 1990 Census data andother OECD statistics on international migration to construct estimates of emigra-tion rates at three education levels (primary, secondary and tertiary schooling) forabout 60 developing countries.2 The Carrington-Detragiache (henceforth CD) esti-mates, however, su¤er from four main shortcomings. First, CD assumed for eachcountry that the skill composition of its emigration to non-US OECD countries isidentical to that of its emigration to the US; for example, Nigerian immigrants inthe UK are assumed to be distributed across educational categories in the same wayas Nigerian immigrants in the US. Consequently, the CD estimates are not reliablefor countries for which the US is not the main destination. Second, at the timeCD conducted their study, the OECD immigration data (regarding the EU, Japan,Switzerland, New Zealand...) did not allow for a full decomposition of the immigrants’origin-mix; more precisely, the OECD published statistics indicating the country oforigin only for the top 5 or 10 sending countries. For small countries not capturedin these statistics, the …gures reported in the CD data set are therefore biased: the

2See Carrington and Detragiache (1998, 1999). Relying on the same assumptions, Adams (2003)provides estimates for 2000.

5

total number of emigrants is under-estimated, and in some cases one is (mis)led toconclude that 100% their immigrants to the OECD area immigrated to the US; asacknowledged by Carrington and Detragiache, this may approximate the reality forLatin America, but is clearly erroneous, for example, in the case of Africa. Third,the CD data set excludes South-South migration, which may be signi…cant in somecases (e.g., migration to the Gulf States from Arab and Islamic countries, to SouthAfrica from neighboring countries, etc.). Finally, the de…nition of a migrant is simplya foreign-born individual residing in the receiving country; it is therefore impossibleto distinguish between immigrants who were educated at the time of their arrival andthose who acquired education after they settled in the receiving country; for example,Mexican-born individuals who arrived in the US at age 5 or 10 and graduated fromUS high-education institutions later on are counted as highly-skilled immigrants. De-spite these various shortcomings, the CD estimates constitute a …rst and very usefulstep toward building a fully-harmonized data set on migration rates by educationlevels.In an attempt to extend Carrington and Detragiache’s work, Docquier and Mar-

fouk (2004) collected data on the immigration structure by education levels and coun-try of birth from most OECD countries in 1990 and 2000. They use the same method-ology and de…nitions as Carrington and Detragiache (1998),3 but extend their work ina number of ways. They use Census data or speci…c and detailed surveys for nearly allOECD countries: Census data reporting educational levels and countries of birth wereused for 11 countries in 2000 and 8 countries in 1990. Survey data were used for 13European countries. These data give the immigrants’ origin for all source countries,not only for the top 5 or 10 countries, thus preventing an under-representation of smallcountries, as was the case for the CD dataset. In addition, for 24 out of 30 receivingcountries, the data collected also give the exact education structure of immigration,while CD had to extrapolate this from the US data, with the US representing onlyabout 44.1% total immigration to the OECD. On the whole, the estimates are basedon accurate information on the immigration stock per country of origin and educationlevel (primary, secondary, and tertiary schooling) for 88.8% of adult immigrants in1990 and 92.7% of adult immigrants in 2000. Hence, Docquier and Marfouk (2004)provide ”reliable” estimates (i.e., based on precise information for at least 80% of thetotal stock analyzed) for 175 countries in 2000 and 160 countries in 1990.4

Aggregating over countries, the total number of adult immigrants living in theOECD area and aged 25 or more is estimated at 58.5 million in 2000 and 39.8 millionin 1990. Emigration rates by education levels are obtained by comparing the number

3Their estimates are based primarily on national census and register data. Survey data werealso used when census statistics were not available or not su¢ciently detailed. In a limited numberof case where data by country of birth are not available, immigrants are de…ned as residents withforeign citizenship (e.g., Germany, Japan, Italy or Korea).

4At the same reliability rate, Carrington and Detragiache (1998) provided reliable estimates forabout 25 countries.

6

of emigrants to the population from which they are drawn. The latter informationis taken from the Barro and Lee (1993) data set on educational attainments. Atthe world level, 1.66% of the working age population is living in a foreign country.The worldwide average emigration rates amount to 0.94%, 1.64% and 5.47% for low-skill, medium-skill and high-skill workers, respectively. As apparent from Figure 1,emigration rates for high skilled workers are strongly correlated to (but always higherthan) total emigration rates.

F igure 1 . Skilled w orkers' and to tal em igration rates(w ith 2-order polynom ia l trend)

y = - 2 .6 9 3 5 x 2 + 3 .2 0 6 8 x - 0 .0 0 6 4R 2 = 0 .8 3 4

0

0 .1

0 .2

0 .3

0 .4

0 .5

0 .6

0 .7

0 .8

0 .9

1

0 0 .0 5 0 .1 0 .1 5 0 .2 0 .2 5 0 .3 0 .3 5 0 .4 0 .4 5

T o ta l m ig r a tio n r a te

Sk

ille

d m

igra

tio

n r

4 5 ° l in e

Regarding receiving countries, two major trends characterize the evolution of mi-gratory patterns between 1990 and 2000: (i) Quality-selection is more and more im-portant and, (ii) Low-income countries are increasingly a¤ected by the brain drain.The worldwide average rate of emigration for skilled workers has increased by 0.75percentage point over the period 1990 and 2000, against only 0.06 percentage pointfor low-skill workers and 0.41 for medium-skill workers. As shown on Figure 2, mostskilled emigrants choose the USA as destination (…g 2.A)5. More importantly, theshare of skilled workers in the total stock of immigrants has increased in all receiv-ing countries, and more so in Canada and Australia, the two countries which werethe …rst to introduce selective immigration policies o¢cially (…g 2.B). In all OECDcountries, the proportion of skilled immigrants originating from low-income countrieshas increased, especially in North America, with a notable increase of highly-skilledimmigration from Asian countries (…g 2.C).

5In 2000, about 44% of the OECD adult immigrants lived in the US. This proportion jumps to50% for skilled immigrants.

7

Figure 2. Immigration structure in the OECD area

USA

EU

-15

Can

.

Au s

tr.

Oth

e rs

0%

10 %

20 %

30 %

4 0 %

5 0 %

6 0 %

A. D est ination of skilled im m igrants in p ercent o f the O ECD

Ye ar 1990 Y ear 200 0

USA

EU-1

5

Can

.

Aus

tr.

OEC

D

0%

1 0%

20 %

30 %

40 %

5 0 %

6 0 %

B. Skil led imm igrants in percen t o f the im m igrat ion stock

V aria tio n 2000-1990 Ye ar 1990 Y ear 2000

USA

EU-1

5

Can.

Aust

r.

OEC

D

0%

5%

10 %

15 %

20%

25%

30%

C. Percentage o f sk illed im migrants from low income countries

V aria tion 2000-1990 Y ear 199 0 Y ear 2000

8

Regarding sending countries, the intensity of the brain drain di¤ers if it is mea-sured in absolute or relative terms. Table 1 provides the 30 most a¤ected and leasta¤ected countries in 2000. Only countries with a population higher than 4 million areincluded. Looking at absolute numbers the Philippines, India, China, Mexico but alsothe United Kingdom, Germany, and Canada appear as the major sending countries.When looking at relative numbers (i.e., in proportion of the local educated labor forceat origin), skilled emigration appears to be particularly strong in Central Americaand the Paci…c Region, with …gures higher than 80% in countries such as Guyana,Jamaica, or Suriname.

Table 1. Emigration (stocks or rates) of skilled workers in selected countries(excluding countries with population < 4 millions)

Em igration stock in 2000 Emigration rate in 2000 Em igration rate in 200030 largest stocks 30 highest rates 30 lowest rates

1 U nited K ingdom 1 542 011 H aiti 81.6% Sw eden 4 .4%2 P hilipp ines 1 260 879 Somalia 58.6% Egypt 4 .2%3 India 1 021 613 G hana 42.9% China 4 .2%4 G ermany 1 016 007 M ozambique 42.0% India 4 .2%5 China 906 337 Sierra Leone 41.0% M oldova 4 .2%6 M exico 901 347 V ietnam 39.0% France 3 .9%7 Canada 566 833 N igeria 36.1% Libya 3 .8%8 I taly 470 331 M adagascar 36.0% Burma (M yanmar) 3 .4%9 V ietnam 446 895 El S alvador 31.5% V enezuela 3 .3%

10 U nited States 428 078 N icaragua 30.9% Brazil 3 .3%11 K orea, North 422 518 Lebanon 29.7% Burkina F aso 3 .3%12 K orea, South 384 497 Croatia 29.4% Belarus 3 .0%13 P oland 379 266 Cuba 28.9% N epal 2 .7%14 Cuba 336 419 H ong Kong 28.7% G eorgia 2 .6%15 J apan 331 892 Papua N ew G uinea 28.2% Azerbaijan 2 .6%16 F rance 301 717 Sri Lanka 27.5% Spain 2 .6%17 I ran 282 587 K enya 26.3% Argentina 2 .5%18 T aiw an 280 710 Angola 25.6% Australia 2 .3%19 Russian Federation 263 041 Senegal 24.1% P araguay 2 .3%20 J amaica 260 850 H onduras 21.8% Thailand 2 .2%21 H ong K ong 254 805 D ominican Repub lic 21.7% Indonesia 2 .0%22 Brazil 254 467 U ganda 21.6% Japan 1 .5%23 N etherlands 240 494 G uatemala 21.5% Rus sian F ederation 1 .3%24 U kraine 237 395 Burund i 19.9% K azakhstan 1 .1%25 Colombia 232 596 Rwanda 19.0% U zbek is tan 1 .0%26 I reland 220 545 Serbia and M ontenegro 17.4% K yrgyzs tan 0 .7%27 Romania 217 198 Ethiop ia 17.0% Saudi Arabia 0 .7%28 P eru 183 915 U nited Kingdom 16.7% Tajikistan 0 .7%29 P akistan 174 884 Tanzania 15.8% U nited States 0 .5%30 N ew Zealand 172 582 Slovakia 15.3% Turkmenistan 0 .1%

Despite a strong within-group heterogeneity, Figure 3 illustrates interesting di¤er-ences between groups. Average migration rates are computed by region, by incomegroup (for the four categories distinguished by the World Bank) and and by country-size groups (above 25 million, from 10 to 25 million, from 4 to 10 million and below4 million). Regarding the regional distribution (see …gure 3.A), the most a¤ected

9

regions are Africa and Europe6. The Paci…c region and Asia exhibit intermediaterates, and the …gures for American countries as a whole are small.7 The regional dis-parities are extremely stable between 1990 and 2000. Regarding income groups (see…gure 3.B), it is worth noticing that the highest rates are observed in middle-incomecountries. High-income countries (less incentive to emigrate) and low-income coun-tries (where liquidity constraints are more binding) exhibit the lowest rates. Between1990 and 2000, the situation clearly improved in lower-middle-income countries anddeteriorated in low-income countries. Finally, regarding country-size groups, thereis a clear decreasing relationship between the emigration rate and the country size.Disparities are also extremely stable over time.Spatial clusters may also be distinguished, as apparent from the world map in

Figure 4. In North and South America, Western and Eastern Europe, South-Centraland Eastern Asia, North and South Africa, brain drain is rather small. On thecontrary, brain drain has a very important impact on the human capital structure ofthe labor force in the Caribbean, Central America, South-Eastern Asia, Western andEastern Africa. The proximity with the USA is an obvious key factor explaining thesituation of the Caribbean and Central America. Countries from South Eastern Asia(Vietnam, Philippines, Laos) have a more diverse portfolio of destinations as they sendmigrants to the USA, Canada and Australia. For African countries, colonial ties withEuropean countries such as the UK (Gambia, Ghana, Somalia), France (Mauritania,Djibouti), or Portugal (Angola, Cape Verde, Mozambique) are key determinants ofmigrants’ locations.

6Migration between European countries is counted.7The American average rate is clearly driven by the US.

10

Figure 3. Skilled workers emigration rates by country group

LargeUpp er

M idd leLow erM iddle Low

1990

20000.000

0.050

0.10 0

0.150

0. 200

0.250

C. Average ra te by country size

1 990 2000

H ighU pperM iddle

L owe rMiddle

L ow

199 0

20000.000

0.020

0.040

0 .060

0.080

0.100

0.120

B. Average ra te by incom e grou p

1990 2000

Am erica Europe Africa Asia Ocean ia

1990

20000.000

0 .020

0.040

0 .060

0.080

0 .100

A. Average rate by region

1990 2000

11

Figure 4. Emigration rate of the highly skilled - world map

We now turn to the economic theories that have analyzed the determinants andconsequences of international skilled-migration ‡ows for developing countries.

3 Theory and evidenceThis section provides an overview of the theoretical and empirical literature on theconsequences of highly-skilled emigration for source countries. This issue has givenrise to a large body of research since the late 1960s, with the early literature generallysupporting the view that the brain drain is detrimental to those left behind and themore recent literature providing a more balanced view. We …rst present the generalset-up, reformulate the results of early contributions within this framework, and thenintroduce the various channels emphasized in later research. We also present theexisting evidence on each particular channel.

3.1 The model

Consider a stylized small open economy populated by two-period lived individuals. Ateach period, a composite good is produced according to a Cobb-Douglas technology,Yt = AtK

1¡®t L®t . For simplicity, the stock of capital, Kt; is assumed to be composed

of foreign investments only (no domestic savings), and the labor supply, Lt; sumsup skilled and unskilled labor. Normalizing the number of e¢ciency units o¤ered byan unskilled individual to 1, a skilled individual o¤ers h > 1 such units. The scalefactor is time-variable and can be positively related to the economy-wide averagelevel of human capital of the adults remaining in the country, Ht; itself a functionof the proportion of skilled workers in that generation, Pt: Therefore, we have Ht =1+Pt(h¡1), with Pt the share of skilled workers and h > 1 their relative productivity,and At = A(Ht);with A

0 ¸ 0 .8This latter mechanism allows us to introduce the spillover e¤ects associated to hu-

man capital formation. The international mobility of capital is such that the marginalproductivity of capital equals to the world interest rate (r¤) plus a risk premium (¼t)associated to internal factors such as risk, political instability, corruption, individualfreedom, etc. Hence, the domestic wage rate per e¢ciency unit of labor is given by:

wt = ®

·1¡ ®r¤ + ¼t

¸ 1¡®®

[A(Ht)]1® ´ w(¼t; Ht)

with the derivatives w01 < 0 and w

02 > 0.

When young, people are o¤ered the choice between working as unskilled workersor devoting part of their time to education. There is a single education program, thecost of which is proportional to the domestic wage rate wt. However, individuals are

8We will later restrict the values of h to ]1; 2[ to obtain interior solution when we assume anuniform distribution of abilities.

13

heterogenous in the ability to learn and may therefore be characterized by di¤erenteducation costs, with high-ability individuals incurring a lower cost. The cost of ed-ucation for a type-c agent is denoted by cwt; with c distributed on [0; 1] accordingto the cumulative distribution F (c). When adult, skilled (educated) and unskilledagents work full-time, with education enhancing one’s productivity and, thus, one’sincome, by an exogenous skill premium h 2 [1; 2].9 Utility is linear in consumptionand there is no time-discount rate. There is no domestic saving so that the stock ofcapital is totally owned by foreign investors.Without migration, the lifetime income for an uneducated agent is wt+wt+1. By

contrast, the lifetime income for an educated agent is wt ¡ cwt + wt+1h. Clearly,education is worthwhile for individuals whose education cost is lower than a criticalvalue. At the steady state (wt+1 = wt), the condition for investing in education in aneconomy with no migration (henceforth denoted by the subscript n) is:

c < cn ´ h¡ 1:In poor countries, however, liquidity constraints are likely to impact on education

choices. Assume, that the …rst-period consumption cannot be lower than a minimalthreshold, Áwt, which is assumed to be proportional to wages. Hence, an agentwith education cost above cL ´ 1 ¡ Á has no access to education, and the liquidityconstraint may or may not be binding depending on whether cL ? cn:Consequently, the economy-wide average level of human capital of the current

generation of adults may be written as:

Hn = 1 + Pn(h¡ 1)where Pn =Min [F (cn);F (cL)] measures the proportion of educated adults.Let us now examine the impact of skilled migration on the sending economy. In

the model with private education funding, the impact of migration on remainingresidents is related to the way it a¤ects the composition of the labor force. Whatmatters is the e¤ect of migration on the equilibrium wage rate. If the stock of humancapital per worker remaining in the home country decreases (resp. increases), thewage rate also decreases (resp. increases) and there is a loss (a gain) of welfare forthose left behind. For di¤erent reasons, we are aware that the relationship betweenwelfare and skilled migration is likely to depend on other channels which are notmodeled here:

² As we shall argue in the policy discussion (see section 4.2), results can bedi¤erent under public funding of education. Suppose maintaining the sameaverage stock of human capital requires increasing subsidies and taxes, thenuneducated workers (who receive no subsidies) experience a welfare loss due tothe reduction in net wages (despite constant gross wages);

9Given c 2 [0; 1], the restriction h < 2 ensures that the proportion of educated is lower than onewhen c is uniformly distributed.

14

² As developed in section 3.5, a rise in welfare can be obtained with decreasinghuman capital when migration gives rise to international transfers. For the mainrecipients, disposable income can rise despite decreasing wage rate;

² Another possible criticism is that migration is likely to a¤ect welfare throughnon-income channels. Buidling on the idea that individuals obtain pleasure frominteraction with those with whom they share social capital (including norms,language, culture and more), Schi¤ (2002) models the negative externality ofemigration per se for those left behind. Even if empirical evidence on socialcapital and migration is limited, the ”social capital drain” is likely to reducewelfare despite constant income.

To avoid using a welfare criterion, our analysis focuses on the impact of the braindrain on the economic potential of the sending country, summarized by the averagestock of human capital among remaining members.

3.2 The traditional view

The …rst ”modern” economic papers on the impact of highly-skilled migration onsource countries date back to the late 1960s (Grubel and Scott, 1966, Johnson, 1967).Their conclusions were not too pessimistic. For example, Grubel and Scott (1966)argued that the short-term loss to the source country (due to the intragenerationaland …scal externalities outlined above) could well be o¤set in the the long run thanksto various possible feedback e¤ects in the form of remittances, networks e¤ects, orinnovations that may spillover from the host to the origin country. However, suchpossible feedbacks were considered too small to make a di¤erence, and the brain drainliterature of the 1970s focused on its detrimental (short-term) impact.The central conclusion of the early brain drain literature, namely, that the brain

drain is detrimental to the welfare of those left behind, relies on a number of criticalassumptions: (i) Migrants are self-selected among the pool of emigrants, (ii) thereis free mobility and, hence, no uncertainty regarding future migration opportunitiesand, (iii) there is a complete disconnection between emigrants and their country oforigin once they have left (no diaspora e¤ect, no return migration, no remittances).In such conditions, clearly, emigration can only a¤ect negatively the proportion ofeducated in the remaining population, P:Building on the stylized model above, consider that workers now have the possi-

bility to emigrate toward a developed country where, due an exogenous technologicalgap, one unit of human capital is paid w¤ > wt. The wage ratio can be written as!t = w

¤=wt = !(Pt) with !0< 0. Migration involves a cost kw¤ which captures trans-

portation, search, assimilation and psychological costs of leaving one’s home country.Individuals have to choose whether to educate or not (ED or NE), and whether tomigrate or not (MI or NM). The lifetime income associated to each pair of decisions

15

is determined by

U(NE;NM) = wt + wt+1

U(NE;MI) = wt + w¤(1¡ k)

U(ED;NM) = wt ¡ cwt + wt+1hU(ED;MI) = wt ¡ cwt + w¤(h¡ k)

At the steady state, the condition for a self-selection equilibrium to emerge (i.e.,skilled workers only emigrate) is :

!(1¡ k) < 1 < !(1¡ kh)

In this case, migration prospects impact on the critical ability level required forinvesting in education; the condition for investing in education becomes:

c < co ´ !(h¡ k)¡ 1

which is higher than cn = h¡ 1 providing that the self-selection condition holds.There is a great deal of evidence that migration prospects indeed impact on peo-

ple’s decisions to invest in higher education. For example, in their survey on medicaldoctors working in the UK, Kangasniemi et al. (2004) evaluate that the migrationpremium in the medical profession lies between 2 and 4 (in PPP values); about 30% ofIndian doctors surveyed acknowledge that the prospect of emigration a¤ected theire¤ort put into studies; furthermore, the doctors surveyed estimate that migrationprospects a¤ect the e¤ort of about 40% of current medical students in India. Focus-ing on the software industry, Commander et al. (2004) estimate that the migrationpremium for Indian IT workers contemplating emigration to the US lies between 3 to5 (depending on the type of job) in PPP values.10 According to the IOM (2003), theprospects of working abroad have increased the expected return to additional years ofeducation and led many people to invest in more schooling, especially in occupationsin high demand overseas.Migration prospects stimulate domestic enrollment in education but actual emi-

gration deprives the country of its educated citizens, the proportion of educated inthe remaining population falls to zero, and the average level of human capital ofremaining members falls to 1.In the presence of a minimal threshold for consumption, migration can be limited

by an additional liquidity constraint. Liquidity constraints are due to the monetaryfraction of the migration costs only (as psychological costs of leaving and assimilationcosts are incurred only once migration has occured). Let us denote by k

0w¤ < kw¤

10In current $, the migration premium is much larger (higher than 10 for many countries). Manymigrants confess that they were unable to compare earnings on a PPP basis. The expected migrationpremium is likely to lie between the PPP and the current $ values.

16

this monetary component of the migration cost. Agents with education costs abovecM ´ 1 ¡ k0! ¡ Á < cL cannot both educate and migrate. A positive number ofeducated individuals thus remains in the source country when the threshold cM islower than cn. In this case indeed, individuals with personal ability between cM andcn cannot a¤ord paying for both migration and education costs but still have anincentive to invest in education (see case 1 on Figure 5). When cM is higher thancn, however, agents who cannot a¤ord paying for migration costs have no incentiveto educate and all the educated would leave the country at the end of period 1 (seecase 2 on Figure 5).

Figure 5. Brain drain, education choices and liquidity constraint

Case 1. A positive share of educated remains Case 2. All the educated leave U U

cM cn co c cn cM co c

Basically, the central prediction of the traditional view is that once migrationopportunities are introduced, the average level of human capital among remainingresidents decreases. The e¤ect on natives’ income depends on A(Ht), through whichvarious types of externalities can be considered. Building on the idea that the so-cial return to education is higher than its private return, the literature of the 1970sgenerally concluded there was a detrimental e¤ect based on the externality argument(Hamada, 1977, Usher, 1977, Blomqvist, 1986). In a similar spirit, Bhagwati andHamada (1974) developed a model of wage determination in which the departure ofskilled workers also reduces unskilled workers’ expected earnings. The mechanismwhereby skilled workers’ emigration negatively impacts remaining workers’ wages in-volve a mechanism of wage-setting that accounts for ine¢ciences of labor marketsin developing countries. Assume that there are two types of workers (educated anduneducated), and wages for the educated are determined by workers’ unions and in-corporates an element of international emulation (i.e., depend positively on wagesabroad). Once skilled-workers wages are set, unskilled-workers wages follow withsome rule of proportionality. In this setting, skilled migration reduces skilled un-employment, meaning that wage pressures become stronger. While the net e¤ecton skilled employment depends on the elasticity of demand for skilled labor (deter-

17

mining whether the skilled labor wage bill increases or not), this tends to extendunemployment and reduces welfare among the uneducated.Note that Bhagwati and Hamada (1974), as well as McCulloch and Yellen (1977),

take into account the incentive e¤ects of the brain drain on education decisions,with the increase in the expected-wage for skilled workers stimulating human capitalinvestments; they also raise a number of questions regarding optimal public …nancingof education in such a context, an issue that we will deal with in Section 4.Modern theories of endogenous growth have considerably renewed the analysis

of the relations between education, migration and growth. Unsurprisingly, the …rstmodels to address the issue of the brain drain in an endogenous growth frameworkalso emphasized its negative e¤ects (e.g., Miyagiwa, 1991, Haque and Kim, 1995).At the same time, a series of studies have tried to promote the simple idea that oneshould also look at how a given stock of human capital is built up. In particular,it is likely that in the presence of huge inter-country wage di¤erentials, as is thecase between developing and developed countries, the prospect for migration deeplymodi…es the incentive structure faced by developing countries’ residents when mak-ing their education decisions. When migration is temporary or when the educationdecision is made in a context of uncertainty regarding future migration opportunities,a bene…cial brain drain or a brain gain may result for the source country.

3.3 Temporary migration

Let us …rst introduce return migration and temporary visas. As documented in in-ternational reports (e.g., OECD, 1998), most receiving countries have recently madeadmission conditions for candidate immigrants more restrictive. On the one hand, asdetailed in the introduction, selective procedures have been put in place; on the otherhand, most new speci…c immigration programs targeting the educated and skilled aredesigned for temporary immigrants, the general trend being towards an increase inthe share of temporary visas relatively to permanent visas. Assume, therefore, thatcandidate immigrants are allowed to spend only a fraction ° of their working lifein the destination economy. Substituting temporary for permanent visas reinforcesself-selection among migrants: the expected return to education being lowered, fewerpeople will invest in education and only those at the upper-end of the ability dis-tribution will …nd it bene…cial to do so. Obviously, the exact impact on those whowould have invested in education would visas had been permanent visas, depends onthe length of the migration period. In terms of our notations, the lifetime income for

18

educated agents are now given by:11

U(ED;NM) = wt ¡ cwt + wt+1hU(ED;MI) = wt ¡ cwt + °w¤h+ (1¡ °)hwt+1 ¡ kw¤

At the steady state, emigration is optimal for skilled workers when the followingcondition holds:

°h(! ¡ 1) > k!If the latter condition does not hold, migration prospects have no e¤ect on human

capital formation. If it holds, the perspective of temporary migration stimulateshuman capital investments.Without liquidity constraints, the condition for investing in education becomes:

c < c° ´ °(! ¡ 1)h+ h¡ 1¡ k! if °h(! ¡ 1) > k!< cn ´ h¡ 1 if not

In the …rst alternative, and assuming a uniform distribution of abilities, the pro-portion of educated workers in the country becomes:

P° =(1¡ °)c°1¡ °c° :

Graphically, the case of temporary migration is similar to the case of permanentmigration, except that the incentive e¤ect is propotional to °. Nevertheless, themajor di¤erence is that, by contrast to the case with permanent visas, the incentivee¤ect partly bene…ts to the sending country. Indeed, the probability P° can be loweror higher than Pn. Formally, a possibility of ”bene…cial brain drain” emerges if thederivative of P° with respect to ° is positive for low values of °, i.e. a value suchthat skilled workers start opting for migration (° = k!

h(!¡1)). We obtain·@P°@°

¸°h(w¡1)=k!

=(h¡ 1)(h¡ 2) + h(! ¡ 1)¡ k!

[1¡ °(h¡ 1)]2 Q 0

If this derivative is positive, there is an interval of ° for which the temporarymigration of skilled workers can stimulate the share of educated workers in the sourcecountry and, in turn, the economy-wide average level of human capital.12 Severalelements are likely to mitigate this result, however. First, liquidity constraints are

11Note that for simplicity we assume migration costs to be identical as in the case of permanentmigration. This could be justi…ed by assuming that higher transportation costs (since people nowtravel both ways) strictly compensate for reduced psychological costs, or that the latter are incurredat the …rst years following immigration. Alternatively, we could assume that in the case of atemporary migration, people incur a migration cost of k0 + °(k ¡ k0) ´ k00.12Clearly, for ° = 1, the traditional view applies: the e¤ect of brain drain is unambiguously

detrimental.

19

likely to limit the size of the incentive e¤ect. If c° > cL, some agents have no access toeducation in spite of the fact that education is optimal, which reduced the likelihood ofa bene…cial brain drain. Similarly, if liquidity constraints restrict migration prospects,the incentive e¤ect is thereby weakened. In the particular case where cM > cn, thenumber of individuals engaging in education is constant and temporary migrationreduces the share of educated workers.Dos Santos and Postel-Vinay (2003) argue that a bene…cial brain drain could

emerge even if the share of educated workers decreases. This is shown in a settingwhere growth is exogenous at destination and endogenous at origin, with the soleengine of growth there being knowledge accumulation embodied in migrants returningfrom the more advanced country. Their caveat relies on knoweldge di¤usion, that is,on the idea that the more advance technology spillovers to the developing country as itis in a way carried out by returning migrants. To the extent that returnees contributeto the di¤usion of the more advanced technology they experienced abroad, emigrants’return is therefore a potential source of growth for their home country. This meansthat return migrants come back with a productivity gain, £h > h, which stimulatesthe average level of human capital. The average stock of human capital then becomes:

H = 1 + P°(£h¡ 1)which must be compared to the case of no migration, H = 1 + Pn(h¡ 1).In a companion paper, Dos Santos and Postel-Vinay (2004) show that a shift in

immigration policy, with an increase in the share of temporary visas, may bene…t tothe sending countries of educated migrants. Two e¤ects of the proposed shift in immi-gration policy are described: a decrease in the incentives to acquire education, whichreduces the pre-migration stock of human capital at origin, and a higher proportionof returnees among emigrants, which increases the country’s stock of knowledge, acomplement of human capital. Their paper derives the theoretical conditions requiredfor an overall positive e¤ect to occur.Using a di¤erent perspective, Stark et al. (1997) elaborate on the possibility of a

brain gain associated with a brain drain in a context of imperfect information withreturn migration. In their setting, workers’ productivity is revealed at destinationonly after a certain period of time during which people are paid according to theaverage productivity of their group. Some relatively low-skill workers will therefore…nd it bene…cial to invest in education so as to migrate and be pooled at destinationwith high-skill workers; once individuals’ ability are revealed, the low-skill workersreturn to their home country, which may therefore bene…t from their educationalinvestments.There is limited evidence that return migration is signi…cant among the highly-

skilled, or that skilled returnees largely contribute to technology di¤usion. We knowthat in general, return migration is characterized by negative self-selection (Borjasand Bradsberg, 1996) and is seldom among the highly skilled unless sustained growthpreceded return. For example, less than a …fth of Taiwanese PhDs who graduated

20

from US universities in the 1970s in the …elds of Science and Engineering returnedto Taiwan (Kwok and Leland, 1982) or Korea, a proportion that rose to about onehalf to two-thirds in the course of the 1990s, after two decades of impressive growthin these countries. Is it due to the economic boom at origin or to changes in theimmigration policy at destination? Recent evidence is quite mitigated.On the one hand, the …gures for Chinese and Indian PhDs graduating from US

universities in the same …elds during the period 1990-99 are fairly identical to whatthey were for Taiwan or Korea 20 years ago (stay rates of 87% and 82%, respectively)(OECD, 2002). This would seem to be con…rmed by a recent survey which showsthat in the Hsinchu Science Park in Taipei, a large fraction of companies have beenstarted by returnees from the USA (Luo and Wang, 2001). In the case of India,Saxeenian (2001) shows that despite the quick rise of the Indian software industry,only a fraction of Indian engineers in Bangalore are returnees. According to thesepapers, return skilled migration appears relatively limited, however, and is often morea consequence than a trigger of growth.On the other hand, a more recent and comprehensive survey of India’s software

industry was more optimistic and con…rmed the presence of network e¤ects and theimportance of temporary mobility (strong evidence of a brain exchange or a braincirculation), with 30-40% of the higher-level employees having relevant work experi-ence in a developed country (Commander et al., 2004). In their survey on medicaldoctors working in the UK, Kangasmieni et al (2004) found that ”many” intend toreturn after completing their training.

3.4 Uncertainty

Before 1965, the US immigration policy was based on country-speci…c quotas. Thisquota system was abolished but various types of requirements and restrictions im-posed by the US and other countries’ immigration authorities render the migrationdecision very uncertain. Implicit or explicit size-quotas are e¤ectively in place, andreceiving an immigration visa, whether temporary or permanent, requires being ina close relationship either with relatives or employers who must then demonstratethat the migrant’s skills can hardly be found among native workers. Moreover, insome countries, point-systems are used to evaluate the potential contribution of im-migrants to the host economy. This means that at all stages of the immigrationprocess, there is a probability that the migration project will have to be postponedor abandonned. Individuals engaging in education investments with the prospect ofmigration must therefore factor in this uncertainty, creating the possibility of a netgain for the source country. The conditions required for this possibility to materializehave been the subject of a number of theoretical contributions (Mountford, 1997,Stark et al., 1998, Vidal, 1998, Beine et al., 2001).To account for this within our general framework, assume that the probability

of migration depends solely on the achievement of a given educational requirement,

21

which is observable, and not on individuals’ ability, which is not perfectly observable(i.e., migrants are assumed to be randomly selected among those who satisfy somekind of prerequisite with informational content regarding their ability - in our case,education).13 The model with uncertainty looks like an out-selection model wherereceiving countries accept a fraction p > 0 of skilled candidates and reject all unskilledapplications. Alternatively, it can be considered as a model of self-selection in whichunskilled workers have no incentive to emigrate while skilled migration is optimal butlimited.Assume moreover that the subjective probability of getting a visa, as seen by a

potential migrant, equals the proportion of the educated who e¤ectively emigratedwithin the previous generation (i.e., at the steady-state). Denoting by p this proba-bility, the lifetime income for educated agents is now given by

U(ED;NM) = wt ¡ cwt + wt+1hU(ED;MI) = wt ¡ cwt + pw¤h+ (1¡ p)hwt+1 ¡ pkw¤

Uncertainty and return migration induce similar e¤ects on the expected return toeducation, which is lowered in both cases by contrast to the case of certain perma-nent migration. However, several di¤erences are worth noticing. First, the incentivemechanism here operates even for low values of p (remember that an incentive ef-fect was obtaining with temporary migration only for ° ¸ k!

h(!¡1)). Second, even forp = °, uncertainty generates more incentives to educate than temporary migration.The reason for this is straightforward and has to do with the fact that uncertaintyreduces expected migration costs. However, uncertainty per se cannot be seen as asource of knowledge di¤usion.At the steady state, the condition for skilled migration being optimal is the same

as under certainty (i.e., 1 < !(1 ¡ kh)); but now education is worthwhile for people

for whom:

c < cp ´ h¡ 1 + ph·!(1¡ k

h)¡ 1

¸Clearly, we have cp = cn when p = 0 and cp = co when p = 1.As in the case of temporary migration, there is a possibility of bene…cial brain

drain for the sending country partly to the incentive e¤ect. The proportion of edu-cated workers in the country becomes Pp =

(1¡p)cp1¡pcp . This proportion Pp can be lower

or higher than Pn. A bene…cial brain drain can be obtained for some ranges of p,

13Our simpli…ed model assumes homogenous skill within educational groups. The size of theincentive e¤ect would be di¤erent with heterogenous skills (see Commander et al, 2002). In reality,immigration authorities may be combining education with other selection devices such as tests ofIQ or host-country language ‡uency. Would IQ be a perfect signal of ability and the only criterionretained, migration could only be detrimental to human capital formation at home. Still, and to theextent that IQ or other tests are imperfect signals of ability, an incentive e¤ect exist for intermediateskilled workers (the probability to emigrate potentially increases with ability).

22

providing that the derivative of Pp with respect to p is positive at p = 0. We obtain:·@Pp@p

¸p=0

= (h¡ 1)(h¡ 2) + h(! ¡ 1)¡ k! Q 0

Note that this derivative corresponds to the numerator inh@P°@°

i°h(!¡1)=k!

.

As in previous cases, liquidity constraints are likely to lower the size of the incen-tive e¤ect. If cp > cL, the incentive e¤ect will be limited to agents with educationcosts comprised between cn and cL. A similar constraint applies if cp > cM .What is the empirical evidence on this ”prospect” channel? To the best of

our knowledge, the …rst study to attempt at estimating the growth e¤ects of thebrain drain using cross-country comparisons is that of Beine, Docquier and Rapoport(2001); in a cross-section of 37 developing countries, and after controlling for remit-tances, they found that migration prospects have a positive and signi…cant impacton human capital formation at origin, especially for countries with low initial GDPper capita levels. This was a …rst but imperfect try since they used gross migrationrates as a proxy measure for the brain drain due to the lack of comparative data oninternational migration by education levels.In a subsequent study, Beine et al. (2003) then used the Carrington-Detragiache

estimates of emigration rates for the highest (tertiary) education as their measure ofbrain drain; again, they found a positive and highly signi…cant e¤ect of migrationprospects on human capital formation, this time in a cross-section of 50 developingcountries. They also computed country speci…c e¤ects, with the following results.First, countries that experienced a positive growth e¤ect (the ’winners’) generallycombined low levels of human capital and low migration rates, whereas the ’losers’were typically characterized by high migration rates and/or high enrollment rates inhigher education (this is quite intuitive, since in this case most migrants are picked upfrom a stock of people that would have engaged in education even without contemplat-ing emigration). Second, they showed that except for extreme cases such as Guyanaand Jamaica, the growth e¤ects of the brain drain were relatively limited: around plusor minus a maximum of 0.2% in terms of annual GDP per capita growth; this is notnegligible, however, in a dynamic perspective. Finally, it was also striking that whilethere were more losers than winners, the latter included the largest countries in termsof demographic size (China, India, Indonesia, Brazil) and represented more than 80%of the total population of the sample. For the most part, these results are apparenton Figure 6; incidentaly, it may be seen from the …gure that the within-country resultpredicted by the theory outlined above (i.e., that some migration should be good aslong as it is not excessive) is what comes out at the cross-country level apparent onFigure 6. The X-axis gives the Carrington-Detragiache migration rates for the highlyeducated and the Y-axis gives the net growth e¤ect of the brain drain as computedby Beine et al. (2003). The variability across countries at given migration rates isdue to the impact of other right hand side variables, and the curve itself is adjustedusing a second-order polynomial.

23

Figure 6. Brain drain and LDC's growth(with 2nd order polynomial trend)

Egypt

South Korea

Dominican Rep Nicaragua

HondurasGuatemala

Mexico

Philippines

Costa Rica

ArgentinaChile

BoliviaUruguayPeruEcuador

Brazil Colombia

Thailand

IndonesiaPakistan

China

India

Venezuela

Paraguay

-0.04

-0.02

0.00

0.02

0.04

0 2 4 6 8 10 12 14 16 18 20

Emigration rate of tertiary educated workers (in %)

Net

effe

ct o

n an

nual

GD

P gr

owth

rate

(in

%)

3.5 Remittances

Migrants’ remittances constitute another channel through which the brain drain maygenerate positive indirect e¤ects for source countries. It is well documented thatworkers’ remittances often make a signi…cant contribution to GNP and are a majorsource of income in many developing countries. Remittances impinge on households’decisions in terms of labor supply, investment, education (Hanson and Woodru¤,2002, Cox Edwards and Ureta, 2003), migration, occupational choice, and fertility,with potentially important aggregated e¤ects. This is especially the case in poorcountries where capital market imperfections (liquidity constraints) reduce the set ofoptions available to members of low-income classes.The literature on migrants’ remittances shows that the two main motivations to

remit are altruism, on the one hand, and exchange, on the other hand.14 Altruismis primarily directed towards one’s immediate family, and then decreases in intensitywith social distance. By contrast, in principle, no such proximity is required in thecase of exchange; the exchange-based theory of remittances posits that remittancessimply ”buy” various types of services such as taking care of the migrant’s assets(e.g., land, cattle) or relatives (children, elderly parents) at home. Such transfers aretypically observed in case of a temporary migration and signal the migrants’ intention

14See Rapoport and Docquier (2004) for a comprehensive survey of the theoretical and empiricalliterature.

24

to return. A particular type of exchange takes place when remittances are de factorepayments of loans used to …nance the migrants’ investments in education and/ormigration, with altruism and social norms and sanctions making the intergenerationalcontract self-enforcing. Hence, it is a priori unclear whether educated migrants wouldremit more than their uneducated compatriots; the former may remit more to meettheir implicit commitment to reimburse the family for funding of education invest-ments, but on the other hand, educated migrants tend to emigrate with their family,on a more permanent basis, and are therefore less likely to remit (or are likely toremit less) than someone moving alone on a temporary basis.In our basic framework with constant marginal utility of income, remittances have

no e¤ect on the marginal cost and gain of education and in‡uence human capitalformation only when liquidity constraints are binding. Let us develop this particularcase when the distribution of abilities is uniform. Without migration, the shareof educated amounts to cL. With migration opportunities, as some educated agentsleave the country, two opposite e¤ects are observed. Initially, the number of educatedremaining in the country falls to cL ¡ cM . If emigrants remit part of their foreignincome, liquidity constraints become less binding for recipients in the source country.The traditional negative e¤ect can therefore, in principle, be compensated by betteraccess to education for those left behind, with the total e¤ect depending on theamounts transferred and on recipients’ location on the ability axis.McCormick andWahba (2000) obtain the result that highly-skilled migration may

bene…t to those left behind in a trade-theoretic model where migration, remittancesand domestic labor-market outcomes are jointly determined and multiple equilib-ria arise, with the high-migration equilibrium pareto-dominating the low-migrationequilibrium. In a setting closer to the one used throughout this paper, Cinar andDocquier (2004) develop a stylized model where skilled emigrants altruistically re-mit part of their earnings to relatives in the source country. They assume that eachremaining resident receives an identical amount of remittances (which depends onthe proportion of migrants, the intercountry wage gap, and the altruistic parameter)and characterize the transition path (i.e., the dynamics of transfers) and the long-runequilibrium of this economy.Assume that at the steady state, this transfer amounts to T . As shown on Figure

7, the e¤ect of remittances is to shift cL and cM to the right. With a uniformdistribution, and given that (cL+T )¡(cM+T ) = cL¡cM , the proportion of educatedand the economy-wide average level of human capital are given by: PT = cL¡cM

1¡cM¡T andHT = 1 + PT (h ¡ 1). A bene…cial brain drain obtains if HT > Hn, that is, ifT > cM(

1cL¡ 1):

25

Figure 7. Brain drain and remittances

cM T+cM cL T+cL cn co c

In other words, for a bene…cial brain drain to obtain through remittances, thetransfer received by each remaining resident must be relatively high so that a largeshare of the population gains access to education. This do not seem to portray theevidence from remittance data available in developing countries. Although remit-tances are generally positively correlated with donors’ incomes, meaning that skilledemigrants are presumably important remitters, the results from household surveysare mixed. At an aggregate level, Faini (2002) shows that migrants’ remittances de-crease with the proportion of skilled individuals among the emigrants and concludesthat ”this result suggests that the negative impact of the brain drain cannot be coun-terbalanced by higher remittances”. This does not imply that remittances by skilledmigrants are negligible, especially if the proportion of temporary migrants increases;for example, Kangasniemi et al (2004) show that nearly half (45%) of Indian medicaldoctors working in the UK remit income to their home country and that remitterstransfer on average 16% of their income.Instead of sending remittances to relatives at home, migrants may return after

they have accumulated savings abroad and use such savings for promoting invest-ment projects (generally in small businesses). There is much evidence that low-skillworkers migrate with the aim of accumulating enough savings so as to access to self-employment and entrepreneurship (e.g., Mesnard, 2004, and Mesnard and Ravallion(2001) for Tunisia, Dustmann and Kirchkamp (2002) for Turkey, Ilahi (1999) for Pak-istan, Woodru¤ and Zenteno (2001) for Mexico, or McCormick and Wahba (2001) forEgypt). The latter study o¤ers additional insights in that it shows that in the caseof literate migrants, both the amount of savings and the migration duration have asigni…cant positive e¤ect on the probability of entrepreneurship upon return, whilethe …rst proposition only holds true for illiterate migrants; this suggests that skill-acquisition may be more important for relatively educated migrants than the needto overcome liquidity constraints. In terms of our notations, increases in the stock ofphysical capital through repatriated savings and acquisition of entrepreneurial skillscan be modeled as a negative shock on the risk premium, ¼; which rises the stock

26

of capital per worker, thus increasing local wages and decreasing the incentives toemigrate.

3.6 Network e¤ects

Our analysis has so far focused on the long run steady state. In the short run, withunanticipated migration, emigration of educated workers is a net loss to the homecountry. As time goes by, however, successive cohorts adapt their education decisionsand the economy-wide average level of education partly (as in Figure 8a) or totallycatches up, with a possible net gain in the long run (as in Figure 8b) thanks tothe various channels detailed above. On the transition path, additional e¤ects arelikely to operate. In particular, there is a large sociological literature emphasizingthe creation of migrants’ networks that facilitate the movement of goods, factors, andideas between the migrants’ host and home countries. In this section we consider twotypes of network e¤ects: networks that facilitate trade, FDI and technology di¤usion,and networks that facilitate further migration.

Figure 8. The dynamic impact of brain drain

Detrimental brain drain Beneficial brain drain Ht Ht

0 t 0 t

An important socio-economic literature has emerge recently to analyze the conse-quences of the constitution of migrants’ network on migration patterns. For example,Massey, Goldring and Durand (1994) outline a cumulative theory of migration, not-ing that the …rst migrants usually come from the middle ranges of the socioeconomichierarchy, and are individuals who have enough resources to absorb the costs andrisks of the trip, but are not so a­uent that foreign labor is unattractive. Familyand friends then draw on ties with these migrants to gain access to employment andassistance in migrating, substantially reducing the costs and risks of movement tothem. This increases the attractiveness and feasibility of migration for additionalmembers, allowing them to migrate and expanding further the set of people withnetwork connections. Migration networks can then be viewed as reducing the cost,and perhaps also increasing the bene…ts of migration (Bauer et al., 2002, Munshi,

27

2003, and McKenzie and Rapoport, 2004, …nd strong evidence of such network ef-fects); in other words, migration incentives become endogenous once network e¤ectsare introduced.Building on this idea, Kanbur and Rapoport (2004) introduce networks e¤ects at

destination in a standard model of selective migration. In the spirit of Carringtonet al. (1996), they assume that migration costs, k, are decreasing with the size ofthe network at destination, that is, with the number of migrants already emigratedabroad. As explained above, the role of migrants’ networks is to di¤use informationon job availability and provide hospitality and help in job search. Hence, past mi-gration progressively raises the expected return to education (net of migration costs)and, therefore, domestic enrollment in education. For a given p or °, this raises theoptimal number of individuals engaging in education and the share of educated work-ers remaining in the country. In this sense, migrant networks have positive e¤ects onhuman capital formation and serve to mitigate the short-run detrimental e¤ects ofthe brain drain.Another type of network e¤ect consists in the creation of business and trade net-

works; such a ”diaspora externality” has long been recognized in the sociologicalliterature and, more recently, by economists in the …eld of international trade (Rauchand Trindade, 2002, Rauch and Casella, 2003). In many instances indeed, and con-trarily to what one would expect in a standard trade-theoretic framework, trade andmigration appear to be complements rather than substitutes (e.g., Gould, 1994). In-terestingly, such a complementarity has been shown to prevail mostly for trade inheterogeneous goods, where ethnic networks help overcoming information problemslinked to the very nature of the goods exchanged (Rauch and Casella, 2002, Rauchand Trindade, 2002). How the relationship of substitutability or complementaritybetween trade and migration is impacted by the skill composition of migration, how-ever, remains unclear. In the same vein, whether FDI and migration are substitutes(as one would expect) or complements remains an unanswered question, althoughmany case-studies suggest that migrants’ networks favor what sociologists have la-belled ”brain circulation” or ”brain exchange” (e.g., Saxeenian, 2001, Arora andGambardella, 2004).In terms of our notations, and as in the case of repatriated savings, such migrant

networks favoring knowledge di¤usion and foreign direct investment can be modeledas a negative shock on the risk premium, ¼; with similar e¤ects on the capital/laborratio and, thus, on domestic wages and incentives to invest in education.

4 Policy implicationsIt follows from the theoretical and empirical survey above that the consequences ofthe brain drain for source countries are ambiguous. Optimal policy responses are thusunclear. In this section, we discuss two important policy issues regarding the sourcecountries, migration and education policies. Our discussion is based on the out-

28

selection model combining liquidity constraints and uncertain migration prospects(similar conclusions would be obtained in the self-selection model by combining liq-uidity constraints and temporary migration). We consider the uniform distributionof ability. Given the developments above, such a model can be summarized by thefollowing set of equations:

cpl ´ Min [cn + ph­; cM ]

Ppl =(1¡ p)cpl1¡ pcpl

where ­ = !(1¡ kh)¡ 1 measures the foreign wage premium net of migration costs,

cn = h¡ 1 is the critical agent in a closed economy, and cM is the critical thresholdlevel of the education cost when liquidity constraints are binding.Intuitively, the premium ­ varies with the level of development of the source

country: it is high in a poor country and closer to unity in a middle-income ortransition country. Liquidity constraints also depend on the level of development, withpeople living in poorer (and more unequal) societies being more credit constrained.Finally, to formalize the role of education in the development process, we sim-

plify the externality assumed in our general setting (At = A(Ht)) and use a simpleassumption regarding the socially optimal level of education. As in Azariadis andDrazen (1990), we assume a threshold externality induced by the average proportionof educated on the economy-wide productivity parameter. This can be written aswt =w for Pt < P and wt = w for any Pt ¸ P . Hence, we consider that P is propor-tion of educated allowing ot exit out of the poverty trap. Moreover, the incentive toemigrate disappears when wt = w.In such a framework, we examine the role of emigration (or immigration) policies

and education policies on the interplay between migration and human capital forma-tion at origin. In doing so, we make two implicit assumptions: (i) increasing H is agood thing and (ii) governments seek to maximize growth. We leave examination ofthese two assumptions for later discussion.

4.1 The optimal rate of migration

As documented in our introduction, international migration is constrained by im-migration policies in receiving countries. Origin countries may also set barriers tothe international mobility of their citizens, although there is an asymmetry here: forreasons beyond the scope of this paper, setting barriers to entry is seen as legitimateby the international community while setting barriers to exit is not.15 Nevertheless,and before we turn to the policy implications of the analysis, it is interesting fromthe perspective of a given source country to ask whether the optimal rate of skilledemigration is positive.15In addition, such barriers to exit are generally not enforceable for the simple reason that you

can catch and expell people once they are in but cannot force them to return once they are out.

29

Assume that liquidity constraints are not binding; then, a bene…cial brain drainobtains if Ppl > cn. This condition can be written as:

­ > ­BB(p) ´ 1

h£ cn(1¡ cn)1¡ p(1¡ cn)

which is represented as the BB-locus in the (­; p) plan. Clearly, BB is increasingin p. The higher the migration rate of skilled workers, the higher the foreign-wagepremium required to enhance human capital formation at home.Similarly, and given the threshold externality introduced above, skilled emigration

as a source of growth may allow the home economy to exit out of the low-wage trapif Ppl > P . This take-o¤ condition can be written as:

­ > ­TT (p) ´ 1

ph£½

P

1¡ p(1¡ P ) ¡ cn¾

which is represented as the TT -locus in the (­; p) plan. Clearly, TT is decreasingin p. Note that ­TT (1) = ­BB(1). The intuition is clear: what matters here isthe expected return to education, which is positively a¤ected by p and ­; hence thepossible tradeo¤ between the two.However, liquidity constraints are binding if cn + ph­ > cM . This condition can

be written as

­ > ­LL(p) ´ cM ¡ cnph

which is represented as the LL-locus in the (­; p) plan. LL is also decreasing in p. Inthe constrained case, a bene…cal brain drain emerges if p < pb ´ cM¡cn

cM (1¡cn) (this criticalmigration rate corresponds to the intersection between ­BB(p) and ­LL(p)). A take-o¤ obtains if p < pg ´ cM¡P

(1¡P )cM (this migration rate corresponds to the intersectionbetween ­TT (p) and ­LL(p)).Figure 9 describes the cases of a poor and an middle-income economy. In poor

countries, the LL curve is always below the TT curve, meaning that skilled emigrationmay bring about a net welfare gain in the long run but not to the point where aneconomic takeo¤ obtains. In the case of a middle-income economy, the LL curveintersects with the TT curve for p = pg 2 [0; 1].The optimal migration rate, p¤, is obtained by maximizing P with respect to p:

the bold line in Figure 9 represents this solution. Below the BB curve, obviously, theoptimal migration probability is zero. Above the BB curve, p¤ is increasing with ­, atleast as long as liquidity constraints are not binding. When p¤(­) intersects with theLL curve, the latter then gives the actual optimal migration probability. Similarly, ifp¤(­) interesects with the TT curve, there are no longer migration incentives and theactual optimal migration probability is then given by this TT curve. It should be clearthat once the TT curve is reached, the corresponding p¤ applies for one generation only

30

since in subsequent periods, the proportion of educated in the remaining populationis such that the threshold externality for education is attained and migration is nota relevant option anymore. To summarize, when incentive e¤ects are limited byliquidity constraints, the optimal migration rate is given by the LL curve, and oncethe economy exits out of the low-wage trap, the optimal migration rate is given bythe TT curve during the time interval required to reach the threshold proportion ofeducated; then, migration comes to an end.Finally, it is noteworthy that at least in principle, a bene…cial brain drain can

be observed for high values of p. All individuals want to educate if cn + ph­ > 1;or equivalently, if ­ > 1¡cn

ph. Such a condition is represented by the II curve, which

interesects with the BB and TT curves at p = 1. Without liquidity constraints, abene…cial brain drain is unambiguously obtained even if p tends to 1. Given that theeconomy lies above the II curve, the small number of remaining agents are all edu-cated. Such a situation is obviously unrealistic since liquidity constraints would limiteducation investments in a ”su¢ciently poor” economy. In terms of our notations,we can expect that liquidity constraints become more binding as we analyze poorereconomies (cM decreases with ­).

Figure 9. Migration rate, human capital and development

Poor economy Transition economy Ω Ω

LL TT II TT LL II

BB

BB

0 pb 1 p 0 pg pb 1 p

Notes. The dashed area corresponds a detrimental beneficial brain drain area. The bold line represents theoptimal emigration probability given the migration premium Ω (inverse representation)

4.2 Education policy

Our analysis relies on the assumption that the cost of tertiary education is entirelyborne privately. However, if the government (mainly because the social return toeducation is higher than its private return) subsidizes human capital investments,

31

public investment in primary, secondary and tertiary schooling are lost when edu-cated workers emigrate. How does the presence of public expenditures in educationa¤ect our results? How should domestic government adjust their education policy toincreasing migration rates?Let us introduce lump-sum taxes and education subsidies in our basic model.

The government levies a tax ¿wh on each remaining resident (educated or not) andprovide a lump-sum transfer µw to each young opting for education. The lifetimeincome for both types of workers are given by

U(NE;NM) = w + w ¡ ¿hwU(ED;MI) = w ¡ cw + µw + pw¤(h¡ k) + (1¡ p)(wh¡ ¿hw)

Agents for which education is optimal are de…ned by

c < cµ ´ h¡ 1 + ph(­+ ¿) + µThe budget constraint of the government can be written as

¿h = G+ cµmµ + #(¿ ; µ)

where G is the amount of general public expenditure per capita, #(¿ ; µ) is a func-tion measuring perception costs (increasing in both arguments), m is the number ofchildren by remaining adult (emigrants leave the country with their children).In an economy without migration, the critical level of ability would be given by

cn ´ h¡1+µn with µn, the amount of subsidy balancing the budget constraint (witha tax ¿n). It is worth noticing that the tax rate has no in‡uence on the optimalinvestment in education since both educated and uneducated agents pay the sameamount of tax. All other things unchanged (i.e. for given µ and ¿), skilled migrationhas now a double positive impact on human capital formation: the return if higherabroad and migration allows escaping paying domestic taxes.Clearly, by stimulating education investment (cµ > cn), brain drain increases the

amount of public expenditures. To keep the budget balanced, the government hasto increase the amount of taxes or decrease the subsidy rate. Let us denote by¢µ the decrease in subsidy decided by the government and by ¢¿ the increase intaxes. The critical level of ability in the presence of migration can be rewritten ascµ ´ cn+ph(­+ ¿n+¢¿)¡¢µ. Compared to an economy without public education,two e¤ects are obtained. One the one hand, the incentive e¤ect of brain drain is largersince migration gives individuals an opportunity to escape …scal pressure. On theother hand, by potentially reducing the amount of subsidy, brain drain can discourageeducation.The conditions for a bene…cial brain drain and for economic take-o¤ become:

­ > ­BB(p) ´ 1

h£ cn(1¡ cn)1¡ p(1¡ cn) +

¢µ

hp¡ (¿n +¢¿ )

32

­ > ­TT (p) ´ 1

ph£½

P

1¡ p(1¡ P )cn¾+¢µ

hp¡ (¿n +¢¿)

In both conditions, the …rst term corresponds to the condition in an economywithout subsidies; the second term captures the cost of reducing subsidies (it makesthe condition for a bene…cial brain drain more restrictive); the third term captures theadditional incentive e¤ect of brain drain (the lower probability to pay taxes makes thecondition for a bene…cial brain drain less restrictive). The analysis can be completedby introducing liquidity constraints. Let us denote by cM the level of ability abovewhich liquidity constraints impede education in an economy without migration. In aneconomy with migration, this level of ability becomes cM¡¢µ (a lower subsidy meansstronger liquidity constraints). Balancing the budget through education subsidiesshifts the ­LL-locus upwards whilst balancing the budget through taxes has no e¤ecton liquidity constraints.Figure 10 compares the consequences of brain drain in an economy without pub-

lic education (dotted lines) and without public education (continuous lines). If thebudget is adjusted through taxes (constant subsidies), the case for a bene…cial braindrain is stronger in the presence of public education. If the budget is adjusted throughsubsidies, the case for a bene…cial brain drain is weaker.

Figure 10. Brain drain and education subsidies

Adjustments through taxes Adjustment through subsidies Ω Ω

LL TT LL TT

BB

BB

0 pb 1 p 0 pg pb 1 p

Notes. The dotted lines depict an economy without public subsidies The continuous lines depict an economy with public subsidies

This results suggests that brain drain should not induce developing countries toreduce education expenditures. It contrasts with the existing literature on the pro-vision of public good and mobility. A study on education policy and brain draincan be found in Justman and Thisse (1997, 2000). They develop a two-region modelwhere education is publicly funded and where skilled workers choose their locationby maximizing utility. They show that the mobility of skilled workers leads to under-investment in public education by local authorities. In the Nash uncoordinated equi-librium, the migration of skills induces deviation from the …rst best optimum. To

33

restore optimality, they recommend sharing the total burden of education betweenthe local public sector and a supranational authority. Without supranational inter-vention, …scal competition reduces the public funding of education.Such results are not directly transposable to the very speci…c case of developing

countries, in which the …scal competition framework hardly appears as relevant. Eco-nomically speaking, developing countries are small. Migration and capital out‡ows(from one isolated sending country) are too small to a¤ect the receiving countries’equilibrium wages and tax rates. Hence, changes in local education policies should notinduce signi…cant political reactions in receiving countries. Stark and Wang (2002)develop an interesting theory where brain drain can help the education system toreach optimality. Selective migration policies acts as a disguized education subsidy.In case of bene…cial brain drain, they demonstrate that a positive probability of mi-gration can enhance welfare of the remaining residents and nudge the economy towardthe social optimum. Hence, less taxes and subsidies are required to reach the socialoptimum.Such a result critically depends on the assumption that the …rst best can be

decentralized. On the contrary, our framework could be seen as a second-best modelwhere maximizing H is the best policy option (implicitly, it means that the maximalattainable level of human capital will be lower than the social optimum). Thatcase seems particularly suited for analyzing developing countries. In this second-bestworld, we …nd out that a tax adjustment is likely to be the best option in case ofdetrimental brain drain.

5 ConclusionThe main conclusion to draw from the above analysis is that for any given developingcountry, the optimal migration rate of its highly educated population is likely to bepositive. Whether the current rate is greater or lower than this optimum is an empir-ical question that must be addressed country by country. This implies that countriesthat would impose restrictions on the international mobility of their educated resi-dents, arguing for example that emigrants’ human capital has been largely publicly…nanced, could in fact decrease the long-run level of their human capital stock. Thisalso suggests that rich countries should not necessarily see themselves as free riding onpoor countries’ educational e¤orts. The di¢culty is then to design quality-selectiveimmigration policies that would address the di¤erentiated e¤ects of the brain drainacross origin countries without distorting too much the whole immigration system;this could be achieved, at least partly, by designing speci…c incentives to return mi-gration to those countries most negatively a¤ected by the brain drain, and promoteinternational cooperation aiming at more brain circulation.On a …nal note, it is important to underline that what seems crucial at this stage

is to extend the empirical research on the growth e¤ects of highly skilled migrationfor source countries. Two main directions are required: case-studies on the sectoral

34

impact of the brain drain, as suggested by Commander, Kangasniemi and Winters(2004); and extension of the cross-country comparisons. In particular, due to datalimitations, existing empirical studies (Beine et al., 2001 and 2003) are based oncross-section regressions, meaning that they neglect the dynamics of migration ratesas well as the dynamics of education levels; in addition, in the absence of a time seriesdimension, it is impossible to control for individual-country e¤ects in the regressionestimates. Given the strong heterogeneity of developing countries in terms of sizes,levels of development, etc., such country-…xed e¤ects are likely to play some rolein the value of the estimates. However, the new estimates computed for 1990 and2000 by Docquier and Marfouk (2004) should make it possible for further empiricalresearch to go part of the way toward addressing these concerns satisfactorily.

6 ReferencesAdams, R. (2003): International migration, remittances and the brain drain: a studyof 24 labor-exporting countries, World Bank Policy Research Working Paper, n. 2972Arora, A. and A. Gambardella (2004): The Globalization of the Software Industry:

Perspectives and Opportunities for Developed and Developing Countries; in A.B.Ja¤e, J. Lerner and S. Stern, eds.: Innovation Policy and the Economy, Vol. 5,Cambridge, Ma.: The MIT Press, forthcoming.Azariadis, C. and A. Drazen (1990): Threshold externalities in economic devel-

opment, Quarterly Journal of Economics, 105, 2: 501-26.Barro, R.J. and J.W. Lee (1993): International Data on Educational Attainment,

Journal of Monetary Economics, 32, 363-394.Bauer, T., Epstein, G.S., Gang, I.N. (2002): Herd e¤ects or migration networks?

The location choice of Mexican immigrants in the US, IZA Discussion Paper No 551,August.Beine, M., F. Docquier and H. Rapoport (2001): Brain Drain and Economic

Growth: Theory and Evidence, Journal of Development Economics, 64, 1: 275-89.Beine, M., F. Docquier and H. Rapoport (2003): Brain Drain and LDCs’ Growth:

Winners and Losers, IZA Discussion Paper No 819, July.Bhagwati, J. N. and K. Hamada (1974): The brain drain, international integration

of markets for professionals and unemployment, Journal of Development Economics,1, 1: 19-42.Blomqvist, A.G. (1986): International migration of educated manpower and social

rates of return to education in LDCs, International Economic Review, 27, 1: 165-74.Borjas, G.J. and B. Bratsberg (1996): Who leaves? The outmigration of the

foreign-born, Review of Economics and Statistics, 78, 1: 165-76.Carrington, W.J. and E. Detragiache (1998): How Big is the Brain Drain?, IMF

Working Paper, No 98.Carrington, W.J. and E. Detragiache (1999): How extensive is the brain drain,

Finance and Development, June: 46-49.

35

Carrington, W.J, E. Detragiache, and T. Vishwanath (1996): Migration with en-dogenous moving costs, American Economic Review, 86, 4: 909-30.Cinar, D. and F. Docquier (2004): Brain drain and remittances: implications for

the source countries, Brussels Economic Review (special issue on skilled migration),forthcoming.Commander, S., M. Kangasniemi et L.A. Winters (2002): The brain drain: curse

or boon? A survey of the literature ”, Paper prepared for the CEPR/NBER Interna-tional Conference on International Trade, Stockholm (24-25 Mai 2002).Commander, S., R. Chanda, M. Kangasmieni and A.L.Winters (2004): Who gains

from skilled migration? Evidence from the software industry, Manuscript, Center forEconomic Performance, London School of Economics.Cox Edwards, A., and M. Ureta (2003): International migration, remittances and

schooling, Journal of Development Economics, forthcoming.Docquier, F. and A. Marfouk (2004): Measuring the international mobility of

skilled workers, 1990-2000 (Release 1.0), Mimeo., The World Bank.Domingues Dos Santos, M. and F. Postel-Vinay (2003): Migration as a source of

growth: the perspective of a developing country, Journal of Population Economics,16, 1: 161-75.Domingues Dos Santos, M. and F. Postel-Vinay (2004): The impact of tempo-

rary migration on human capital accumulation and economic development, BrusselsEconomic Review, forthcoming.Dustmann, C. and O. Kirchkamp (2002): The optimal migration duration and

activity choice after remigration, Journal of Development Economics, 67, 2: 351-72.Faini, R. (2002): Développement, commerce international et migrations, Revue

d’Economie du Développement, 2: 85-116.Fujita, M., P. Krugman and A. Venables (1999): The spatial economy : cities,

regions and international trade, Cambridge, Ma. : MIT Press.Grubel, H. G. and A. Scott (1966): The international ‡ow of human capital,

American Economic Review, 56: 268-74.Hamada, K. (1977): Taxing the brain drain: A global point of view, in J. Bhagwati,

ed.: The New International Order, Cambridge, Mass.: M.I.T. Press.Hanson, G.H. and C. Woodru¤ (2002): Emigration and educational attainment

in Mexico, Mimeo., University of California at San Diego.Haque, N.U. and A. Jahangir (1999): The quality of governance: second-generation

civil reform in Africa, Journal of African Economies, 8: 65-106.Haque, N.U and S.J. Kim (1995): ’Human capital ‡ight’: impact of migration on

income and growth, IMF Sta¤ Papers, 42, 3: 577-607.Ilahi, N. (1999): Return migration and occupational change, Review of Develop-

ment Economics, 3, 2: 170-86.IOM (2003): World migration 2003 - Managing migration, Geneva: International

Organization for Migration.

36

Johnson, H. (1967): Some economic aspects of the brain drain, Pakistan Develop-ment Review, 7, 3: 379-411.Justman, M. and J.F. Thisse (1997): Implications of the mobility of skilled labor

for local public funding of higher education, Economic letters, 55: 409-412.Justman, M. and J.F. Thisse (2000): Local public funding of higher education

when skilled labor is imperfectly mobile, International Tax and Public Finance, 7(3):247-258.Kanbur, R. and H. Rapoport (2004): Migration selectivity and the evolution of

spatial inequality, Mimeo, Cornell University.Kangasmieni, M., A.L. Winters and S. Commander (2004): Is the medical brain

drain bene…cial? Evidence from overseas doctors in the UK, Manuscript, Center forEconomic Performance, London School of Economics.Kwok, V. and H. Leland (1982): An economic model of the brain drain, American

Economic Review, 72, 1: 91-100.Lowell, L.B. and A.M. Findlay (2001): Migration of highly skilled persons from

developing countries: impact and policy responses, Geneva: International LabourO¢ce (Draft Synthesis Report).Massey, D.S., L. Goldring and J. Durand (1994): Continuities in Transnational

Migration: An Analysis of Nineteen Mexican Communities, American Journal ofSociology, 99, 6: 1492-1533.McCormick, B. and J. Wahba (2000): Overseas unemployment and remittances

to a dual economy, Economic Journal, 110: 509-34.McCormick, B. and J. Wahba (2001): Overseas work experience, savings and

entrepreneurship amongst return migrants to LDCs, Scottish Journal of PoliticalEconomy, 48, 2: 164-78.McCullock, R. and J.T. Yellen (1977): Factor mobility, regional development and

the distribution of income, Journal of Political Economy, 85, 1: 79-96.McKenzie, D. and H. Rapoport (2004): Network e¤ects and the dynamics of

migration and inequality: theory and evidence from Mexico, Working Paper No 63,Bureau for Research and Economic Analysis of Development (BREAD), HarvardUniversity, April.Mesnard, A. (2004): Temporary migration and capital market imperfections, Ox-

ford Economic Papers, 56: 242-62.Mesnard, A. and M. Ravallion (2001): Wealth distribution and self-employment

in a developing country, CEPR Discussion Paper DP3026.Miyagiwa, K. (1991): Scale economies in education and the brain drain problem,

International Economic Review, 32, 3: 743-59.Mountford, A. (1997): Can a brain drain be good for growth in the source econ-

omy?, Journal of Development Economics, 53, 2: 287-303.Munshi, K. (2003): Networks in the modern economy: Mexican migrants in the

US labor market, Quarterly Journal of Economics, 118, 2: 549-99.OECD (1998): Trends in international migration, Paris: OECD Editions.

37

OECD (2000): Trends in international migration, Paris: OECD Editions.OECD (2002): Trends in international migration, Paris: OECD Editions.Rapoport, H. and F. Docquier (2004): The economics of migrants’ remittances,

in L.-A. Gerard-Varet, S.C. Kolm and J. Mercier-Ythier, eds.: Handbook of the eco-nomics of giving, reciprocity and altruism, Amsterdam: North-Holland, forthcoming.Rauch, J.E. and V. Trindade (2002): Ethnic chinese networks in international

trade, Review of Economics and Statistics, 84, 1: 116-30.Rauch, J.E., Casella, A. (2003) Overcoming informational barriers to international

resource allocation: Prices and Ties, Economic Journal, 113, 484: 21-42.Saxeenian, A. (2001) Bangalore, the Silicon Valley of India?, CREDPR Working

Paper No 91, Stanford University.Schi¤, M. (2002): Love thy neighbor: trade, migration and social capital, Euro-

pean Journal of Political Economy, 18, 1: 87-107.Stark, O., C. Helmenstein and A. Prskawetz (1997): A brain gain with a brain

drain, Economics Letters, 55: 227-34.Stark, O., C. Helmenstein and A. Prskawetz (1998): Human capital depletion,

human capital formation, and migration: a blessing or a ’curse’?, Economics Letters,60, 3: 363-7.Stark, O. and Y. Wang (2002), Inducing human capital formation: migration as

a substitute for subsidies, Journal Public Economics, 86(1): 29-46.Usher, D. (1977): Public property and the e¤ects of migration upon other residents

of the migrants’ countries of origin and destination, Journal of Political Economy, 85,5: 1001-20.Vidal, J.P. (1998): The e¤ect of emigration on human capital formation, Journal

of Population Economics, 11, 4: 589-600.Woodru¤, C. and R. Zenteno (2001): Remittances and micro-enterprises in Mex-

ico, Mimeo., University of California at San Diego.World Trade Organization (2004): Trade Statistics, Historical Series 1980-2003

(http://www.wto.org/english/res_e/statis_e/statis_e.htm).

38